Choose the smallest useful system
Separate rules, assistants, agents, and human decisions so AI is used only where bounded judgment has a purpose.
40lessons
10modules
$0Free course
↗Open access
What this course changes
Build the service around a real job and a named owner.
A polished demo can hide a weak workflow. This course starts with the operating problem, then works through context, actions, approval, evaluation, launch, commercial scope, and ongoing responsibility.
Design for failure before launch
Map hard paths, permissions, approvals, data authority, evaluation cases, stop conditions, incidents, and human handoff.
Sell a result you can deliver
Use discovery, a proof-based demo, explicit scope, pricing, onboarding, and a controlled pilot instead of hype.
Operate across clients
Standardize the parts that repeat while keeping client policy, data, access, quality, usage, margin, and support visible.
Complete curriculum
Forty lessons from the first workflow to a defensible agency service.
Every lesson produces a decision, artifact, or review rule. The capstone joins them into an offer and pilot plan a skeptical operator can inspect.
Your progress
0 / 40 lessons
Saved on this device.
01Start with the job4 lessons+
- 01What an AI agent is — and is notChoose the smallest useful system for the work instead of calling every automation an agent.
- 02The economics of a useful agentMeasure an agent by the business constraint it improves, not by how impressive the demo looks.
- 03Find the workflow bottleneckMap the real journey from trigger to outcome before trying to automate the visible symptom.
- 04Decide whether AI belongs in the workflowUse an explicit scorecard before investing in an agent build.
02Find a market and package the work4 lessons+
- 05Pick a market without guessingChoose a reachable segment with a repeated operational pain and an identifiable buyer.
- 06Research conversations, not just competitorsTurn customer language and workflow evidence into a useful problem inventory.
- 07Turn a use case into an offerPackage a believable before-and-after workflow with scope, responsibilities, and proof.
- 08Price for delivery and ongoing responsibilityBuild transparent pricing around the work, usage, support, and risk you actually carry.
03Design an agent people can trust4 lessons+
- 09Give the agent one accountable jobWrite an agent contract that makes its purpose, limits, and owner inspectable.
- 10Map the happy path and the hard pathsDesign the normal journey and the exceptions before real customers discover them for you.
- 11Write instructions people can auditTurn policies and SOPs into clear, testable instructions rather than one oversized prompt.
- 12Design tools, approvals, and handoffsGive an agent only the actions and information the job truly needs.
04Build the first useful workflow4 lessons+
- 13Set up the client workspace and source of truthMake ownership, lifecycle, and authoritative records clear before connecting an agent.
- 14Capture intent from forms, funnels, and conversationsTurn a new signal into enough useful context for a person or workflow to act.
- 15Build the next-best-action workflowMake one narrow workflow that qualifies, records context, alerts an owner, and begins approved follow-up.
- 16Make the handoff feel continuousPreserve the customer’s context when a person takes over.
05Serve people over voice and messaging4 lessons+
- 17Design an AI receptionist that helps, not blocksCreate a concise inbound experience that identifies purpose, gathers permitted details, and routes to a person when needed.
- 18Qualify, route, and book with contextAsk only approved questions, use transparent criteria, and preserve a clear human route.
- 19Use messaging for timely, welcome follow-upMake follow-up useful, permission-aware, and easy to stop or hand to a person.
- 20Handle difficult conversationsDesign for uncertainty, complaints, sensitive requests, and emergencies before launch.
06Connect data and systems safely4 lessons+
- 21Build a source-of-truth mapDocument where each fact starts, which system owns it, and how a correction flows back.
- 22Connect APIs and webhooks with guardrailsTreat integrations as operational systems with retries, audit trails, and safe failure states.
- 23Build a knowledge base the agent can citeCurate approved, current information with owners and retrieval boundaries.
- 24Protect data and permissionsApply least privilege, tenant separation, retention choices, and regular access review.
07Test, monitor, and improve4 lessons+
- 25Build an evaluation set before launchTest representative normal, edge, adversarial, and failure cases against expected outcomes.
- 26Measure the outcome, not only the conversationTrack completion, quality, handoff, correction, and customer experience together.
- 27Operate with layered safeguardsCombine permission, input, tool, approval, logging, and human safeguards instead of relying on one prompt.
- 28Run incidents, change control, and continuous improvementRelease changes deliberately and learn from failures without hiding them.
08Sell the implementation honestly4 lessons+
- 29Build a proof-based demoShow one believable workflow with assumptions, fallback, and visible operational value.
- 30Find conversations without spamCreate an ethical acquisition system using relevance, permission, proof, referrals, partnerships, and useful education.
- 31Run discovery that produces an architectureUse discovery to understand the current workflow, decision rights, systems, risks, and measurement before proposing a build.
- 32Sell scope, not hypePresent the delivery phases, shared responsibilities, exclusions, risk controls, and success measures plainly.
09Deliver client and white-label work4 lessons+
- 33Onboard a client without losing contextCollect the people, evidence, access, policy, baseline, and approvals required for a controlled build.
- 34Deliver a controlled pilotLaunch with a bounded audience, stop conditions, daily review, and a clear decision point.
- 35Build a white-label client experienceCreate branded workspaces with separated permissions, clear support, and honest service expectations.
- 36Run client reviews that keep the work usefulUse evidence-led reviews to discuss outcomes, exceptions, changes, and next decisions.
10Scale a responsible agency4 lessons+
- 37Standardize the 20 percent that repeatsTurn recurring delivery work into reusable templates while preserving client-specific judgement.
- 38Manage margins, capacity, and usageUnderstand labor, provider usage, support, and service limits before scaling delivery.
- 39Build a delivery team and partner networkDefine roles and access so strategy, implementation, QA, support, and specialists can work without broad production permissions.
- 40Capstone: launch and defend an agent serviceBring the course together in a complete, evidence-led offer, architecture, pilot, and operating plan.
Course project
Launch and defend one agent service.
The capstone is not a pitch deck. It is the operating case for a narrow service: who it helps, what it may do, what can go wrong, how you will know, and who owns every decision.
- A workflow and agent job contract
- A source-of-truth, tool, permission, and handoff map
- An evaluation set with success, quality, and safety checks
- A scoped offer, price, onboarding, and controlled pilot
- A 90-day operating plan for reviews, incidents, usage, and change
Questions, answered
What to know before you begin
Do I need to be a developer?
No. The course explains the system in operational terms and includes tools, APIs, webhooks, knowledge, evaluation, and security. Technical readers can go deeper, but the decisions are written for founders, operators, and agencies too.
What counts as an AI agent in this course?
A bounded system that can use context to choose among allowed actions for one accountable job, then report what it did. A fixed rule is automation; a draft that a person approves is an assistant.
Does the course cover voice agents?
Yes. One module covers reception, qualification, routing, booking, messaging, difficult conversations, consent, urgency, sensitive requests, and human escalation.
How does it handle safety and security?
The course covers permissions, source authority, tool allowlists, approvals, data handling, evaluation sets, adversarial and failure cases, layered safeguards, monitoring, incidents, rollback, and change control.
Will this teach me to sell AI-agent services?
Yes. It covers market selection, research, offer design, pricing, proof-based demos, ethical acquisition, discovery, proposals, onboarding, pilots, white-label delivery, reviews, margin, capacity, and teams.
What do I build for the capstone?
A defensible agent service with a narrow workflow, owner, context and tools, permissions, evaluation set, safeguards, commercial scope, pilot plan, and 90-day operating review.
Build the operating layer around the agent.
Use Spacebrain to connect intake, CRM context, calls, messages, routing, tasks, automations, approvals, and reporting.